Content uploaded by Baidaa A. Atya
Author content
All content in this area was uploaded by Baidaa A. Atya on Jul 01, 2020
Content may be subject to copyright.
2019 International Conference of Computer and Applied Sciences (CAS2019), Baghdad, Iraq
Predict the relationship between autism and the use
of smart devices by children in the coming years
using neural networks
Baidaa A. Atya
Computer science department
Education College, Al-mostansiryah University Baghdad, Iraq
Email:dr.baidaa.a@uomostansiryah.edu.iq
Omran T. Ali
Computer science department
Education College, Al-mostansiryah University
Baghdad, Iraq
Email:omrantaha524@gmail.com
xxxxxxxxxxxxxxxxxxxxxxxxx ©2019 IEEE
Abstract— Recently, autism has been particularly widespread
among children. It is no secret that the use of smart devices have
a significant impact on the emergence of this disease directly or
indirectly. In this research the disease was studied in two ways:
First, identify the symptoms of the disease through a
questionnaire that shows how to use smart devices by children
and identify the disease by answering it.
Using an artificial neural network, you can determine
whether a person is infected or not.
The second part is to determine the statistics of the disease in
previous years and through the proportion extracted by the
number of infected through the questionnaire can predict the
number of infected in subsequent years and also using artificial
neural networks.
Through the results we have been able to know the
proportion of infected in subsequent years and thus can address
this disease and make the use of smart devices are part of the
treatment and not the one that helps the emergence of the
disease. To complete this search using VB language.
Keywords— autism disease, artificial neural networks, develop
backpropagation.
I. INTRODUCTION
Autism Disorder Spectrum is a term referring to
complicated disorders in human brain development, through
which people that have Autism spectrum disorder (ASD) could
be facing issues with communication, touching, studying, and
interaction with others. Based on a report from the Network of
Autism and Developmental Disabilities Monitoring (ADDM)
in the U.S., the percentage of children that have been
diagnosed with the ASD is 1/68, whereas it has been 1/110 in
2009 [1], this paragraph provides a simplified definition of
autism.
ASD which is defined by impairments in communications
and social deficits (“Knaus et al., 2008”) is usually a
neurological disease that has high heredity (“Baird et al.,
2006”) and prevalence (“Chakrabarti and Fombonne, 2001”). It
has been stated that ASD prevalence was raised from 00.67%
in 2000 to 01.47% in 2010 (“Xu et al., 2018”). Which is why,
it is important to diagnose ASD early. On the other hand, the
conventional approaches of diagnostics have been
fundamentally based on the behavior observations and clinical
interviews, increasing the accuracy of diagnosis. There are 2
ways which may be implemented for improving the precision
of the diagnostics. One of those ways is using the neuro-
imaging approach, like the positron emission tomography
(PET; “Pagani et al., 2017”), Electro-encephalogram (EEG;
“Peters et al., 2013”), functional magnetic resonance imaging
(fMRI; “Ren et al., 2014”), and structural magnetic resonance
imaging (sMRI; “Sato et al., 2013”). The particular fMRI
characteristics make it commonly utilized (“Bennett et al.,
2017”). One more way is using approaches of ML that would
be capable of automatically improving the efficiency of the
algorithm according to earlier expertise (“Jordan and Mitchell,
2015”). The neural networks (NNs) are an ML branch that has
been inspired with the human brain and performs effective
pattern identification. NNs have been used with success in
automated classification which is associated with the ASD. For
example, Iidaka (2015) has implemented probabilistic neural
network (PNN) for the classification of patients with ASD and
typical controls (TC), and the precision converges to 90%. Guo
et al. (2017) have presented an innovative feature extraction
based on the deep NNs (DNNs) for the classification of
patients with ASD and TC, and the accuracy was 86.36%.
Heinsfeld et al. (2018) have utilized the deep learning for the
diagnosis of the ASD, and the accuracy was 70%. Heinsfeld et
al. (2018) have utilized deep learning for the classification of
TC and ASD patients, and the precision was 70%. Those
researches fully demonstrate that the single NN accuracy isn’t
high and unstable in diagnosing some diseases [2].
So there are so many researchers studied this disease and
putting a lot of solutions to detect diagnosis of autism spectrum
disorder, but in our search we detect the relation between this
disease with using smart devices in these years and in future.
II. RELATED WORK
Lakhwinder K., Vikas K., 2017 [3], This research has
discussed autism concept, its symptoms and signs,
diagnosing and different technologies that are utilized
to diagnose and treat children with autism. A variety
of ANNs and fuzzy based system have been utilized
in the detection of severity levels of autism. With the
use of the ANNs and fuzzy system it is possible
diagnosing a child whether or not they have autism.
Avishek C., Christopher M., 2018 [4], in this paper it
has utilized a data-set which is associated with autism
screening, which includes 10 personal attributes
and10 behavioral which were efficient to diagnose
cases of ASD from the controls in the behavioral
science. The diagnosis of ASD is an expensive and
time consuming process. The burgeoning cases of the
ASD all over the world mandate the necessity for fast
and inexpensive tool of screening. Their ANN with
the algorithm of Levenberg Marquardt for the
detection of the ASD and examining its predictive
accuracy. Successively, developing a clinical decision
support system for the early characterization of the
ASD.
Nguyen V., Ngo L., 2018 [1], have presented the
elaboration on the approach of the utilization of this
combination for the facilitation of early ASD
diagnosis. The results that have been presented in the
research have shown that this method can potentially
be the fundamental basis of the supporting decision-
making system in ASD research and diagnosis.
III. NEURAL NETWORK
Artificial neurons are computational models depend on the
biological neurons. The concept behind the development of
ANNs was transferring the parallel processing concept to
computers for the sake of taking advantage of the features of
the brain. Neural network system is computation between
engineering and artificial intelligence [5].
There are several NN types:
ANNs can be defined as computational models which are
inspired with the biological NNs, and are utilized for
the approximation of the functions which are unknown in
general. Specifically, they’re inspired by neuron behaviors and
the
electric signals that they carry between inputs (like the shape of
eyes or endings of the nerves in the hand), processing, and
outputs from brain (like reactions to touch, heat, or light). The
way of the semantic communication of the neurons is an on-
going research area.
The majority of the ANNs are only somehow similar to
their more complicated biological equivalents, however they
are quite efficient at their aimed functions (such as
segmentation or classification).
Some of the ANNs are adaptive systems which are utilized
for instance, for modelling environments and populations that
always change.
NNs may be hardware-based (in which neurons are
characterized by physical components) or software-
based (computer models), and can utilize various learning
algorithms and topologies.
The feed-forward NN is the earliest and simplest of he
types, where information moves only from input layer in a
direct way through the hidden layers to output layer with no
cycles/loops. Feed-forward NNs may be produced with
different unit types, like the binary McCulloch–Pitts neurons,
the simplest of which is the perceptron. Continuous neurons,
frequently with sigmoidal activation, are utilized in the back-
propagation context. [6].
Fig.1. the structure of multi-layer perceptron
IV. THE PROPOSED WORK
The idea of the research is to create a questionnaire
consisting of 10 questions for children aged 2 to 10 years.
Once the answers are identified, they are inserted into an
artificial neural network to determine whether the child has
autism. Then, a statistical calculation of the number of
casualties as a result of the use of the devices and compared
with the statistics of previous years, and by entering this
information into the network is predicted the number of injured
in subsequent years. The chart below illustrates the structure of
the proposed work.
Fig.2. the structure of proposed work
There are two parts of this work:-
A. Part one: Network training to predict the rate of disease
for children infected with the disease due to the use of
smart devices
.
In this part we uses a neural network (back
propagation) to compute the ratio of children those
infected with the disease due to the use of smart devices
by using multi questions to determine if the child
Infected or uninfected.
Fig.3. Question to determine if the child infected or
uninfected.
Then these answer enter to network to detect the
ratio of the incidence of this disease.The neural
network has three layers: Input (10 nodes), hidden
(10 nodes) and output (1 node). The equation
below computes the error of network:
p n3
E=1/2 ∑ ∑ (dck-Ock) ≤ є (1)
C=1 K=1
Where:-
dck/ the desire output for k cell and for c state.
Ock/ the actual output for k cell and for c state.
є/ the maximum error.
B. Part two: Prognosis by training the neural network on
the statistics of previous years to prophesy injury rates in
the coming years
In this part In this part, the statistics of the previous years with
the statistics obtained through the first part are entered into the
second neural network to develop the statistics for the
following years and to know the number of infections due to
the use of smart devices. This network consists: (7 nodes of
input), (7 nodes of hidden) and (one output node).
The figure below describes the networks:-
Fig.4. Neural Network structure of part one.
Fig.5.The structure of part two to describe the second
neural network training.
IV.
THE
RESULT
A. Information about samplesthat taken and applied to
the system:-
In this research were taken 25 samples between the ages of
(2-12) years and the system was applied to them by making
them use smart devices such as mobile phones and iPad and
others. After several tests, 7 children were found to have the
disease, those children was chosen randomly .
B. Information about the parameters that used when
applied the part one and two:-
There are many parameters that used when applied neural
networks in this research. The table below describes the effect
of the parameters on the network either in the first or second
part:-
TABLE I. THE PARAMETERS OF NETWORKS
C. The final Result of Neural Network Part1:-
Through the implementation of the neural network and by
entering the data obtained by answering the questionnaire for
each child entered for the test, and after training the neural
network to obtain the proportion of children infected with the
disease as a result of the use of smart devices were reached the
results shown in the following table: -
The Parameters
The Values
Neural
Network
Part1
Neural
Network
Part2
The max generation (max-iteration)
The probability of Alfa
The factor of changing the loop (Beta)
The Total network error value
The numbers of hidden nodes (maxh)
The numbers of collections
The range of initial weights (wg)
30
0.95
2
0.003
10
100
[-0.9,0.9]
15
0.9
2
0,008
15
150
[-0.5,0.5]
TABLE II. THE RESULT OF PART ONE
D. The final Result of Neural Network Part2:-
In this part, the neural network was used to predict the
percentage of children with autism in the coming years, thro
ugh the global statistics obtained and also the proportion of
children infected after being identified in the first part of the
research.
The tables below shows the actual global statistics obtained
by some organizations such as UNICEF [7][8]:
TABLE III. THE ACTUAL GLOBAL STATISTICS1
Statista published the results of a 2017 study of autism rates
in 10 countries around the world, which adopted the number of
cases
for
every
10,000
people.
TABLE IV. THE ACTUAL GLOBAL STATISTICS2(2017)
These statistics used as input data in neural network with
the statistics that obtained from neural network in part one to
training network to guess the final ratio for the coming years.
The table below show the result of part two:-
TABLE V. THE RESULT OF PART TWO
No. of
child
The Results
Generation Effort Cost CPU
Time(sec.)
5 15 75 0.003 0.7
10 20 200 0.00463 1.5
15 30 450 0.007 5
20 35 700 0.00869 7
25 50 1250 0.0105 9
No. Sstatistics
country Number of
injured
1Japan 161
2U.K. 94
3Sweden 72
4Denmark 68
5U.S.A 66
6Canada 65
7Australia y 45
8Brazil 27
9Hong Kong 17
10 Portugal 9
11 Morocco 30
12 Algeria 5
13 Turkey 35
No. Sstatistics
Year Ages Number of injured
1 2000 1-7 years 1 for each 150
2 2002 1-7 years 1 for each 150
3 2004 1- 8 years 1 for each 125
4 2006 1-9 years 1 for each 110
5 2008 1-11 years 1 for each 88
6 2010 1-13 years 1 for each 68
7 2012 1-15 years 1 for each 68
IV. CONCL
UTION
Through the results obtained through the introduction of
data and statistics and the training of two neural networks, the
incidence of the disease was reached for children between (2-
12) years and who use smart devices in the coming years
ranging from (55%) to (70%).
These percentages have emerged from those present and in
the past years and the percentage obtained as a result of tests
carried out on the samples taken.
Therefore, I recommend as a future work or development
of the current work to build modern software systems that help
children with autism spectrum using smart devices, but as a
treatment component and not as a component to help
strengthen the disease.
REFERENCES
[1] V. Nguyen, L. Ngo, "Artificial Neural Network and Fuzzy
Logic Approach to diagnose Autism Spectrum Disorder",
International Research Journal of Engineering and
Technology, Volume: 05 Issue: 06 | June 2018.
[2] B. Xia-an, L. Yingchao, J. Qin, S. Qing, S. Qi and D.
Jianhua , " The Diagnosis of Autism Spectrum Disorder
Based on the Random Neural Network Cluster", Front.
Hum. Neurosci., 26 June 2018.
[3] K. Lakhwinder, K. Vikas, "A REVIEW ON USING
ARTIFICIAL NEURAL NETWORK IN DIAGNOSIS OF
AUTISM SPECTRUM DISORDER", Department of
Computer Science and Engineering, CT Institute of
Engineering Management and Technology, Shahpur, 2017.
[4] C. Avishek, M. Christopher,"Prognosticating Autism
Spectrum Disorder Using Artificial Neural Network:
Levenberg-Marquardt Algorithm ", Arch Clin Biomed Res
2018; 2 (6): 188-197.
[5] J. DeMuro, "What is a neural network?", site:/
https://www.techradar.com/news/what-is-a-neural-
network, 2018.
[6] Y. LeCun (2016). Slides on Deep Learning Online.
[7] https://www.wish.org.qa/wp-
content/uploads/2018/01/Arabic_WISH_Autism_Report_FINAL.pdf
[8] https://www.alaraby.co.uk/society/2018/4/2/ - - - -
- - -
[9] I. L. Cohen, and others, "A neural network approach to the
classification of autism", Journal of Autism and
Developmental Disorders, 1993.
[10] L. James McClelland, "The Basis of Hyperspecificity in Autism:A
Preliminary Suggestion Based on Properties of Neural Nets", Carnegie
Mellon University and the Center for the Neural Basis of Cognition,
2000.
[11]
No. of
child
The Results
Generation Cost CPU
Time(sec.)
161 5 0.0007 0.3
94 8 0.0003463 0.52
72 10 0.0008 0.215
68 15 0.00189 0.1297
66 20 0.003105 2
65 23 0.0053 2.51
45 31 0.00821 2,97
27 35 0.00123 3.23
17 39 0.0031 3.38
942 0.00632 3.52
30 50 0.00841 4.176
560 0.00946 4.271
